1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-11-27 15:16:57 +01:00
youtube-dl/youtube_dl/extractor/telegraaf.py

90 lines
2.9 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
determine_ext,
2019-11-03 22:04:03 +01:00
int_or_none,
parse_iso8601,
try_get,
)
class TelegraafIE(InfoExtractor):
2019-11-03 22:04:03 +01:00
_VALID_URL = r'https?://(?:www\.)?telegraaf\.nl/video/(?P<id>\d+)'
_TEST = {
2019-11-03 22:04:03 +01:00
'url': 'https://www.telegraaf.nl/video/734366489/historisch-scheepswrak-slaat-na-100-jaar-los',
'info_dict': {
2019-11-03 22:04:03 +01:00
'id': 'gaMItuoSeUg2',
'ext': 'mp4',
2019-11-03 22:04:03 +01:00
'title': 'Historisch scheepswrak slaat na 100 jaar los',
'description': 'md5:6f53b7c4f55596722ac24d6c0ec00cfb',
'thumbnail': r're:^https?://.*\.jpg',
'duration': 55,
'timestamp': 1572805527,
'upload_date': '20191103',
},
'params': {
# m3u8 download
'skip_download': True,
},
}
def _real_extract(self, url):
2019-11-03 22:04:03 +01:00
article_id = self._match_id(url)
2019-11-03 22:04:03 +01:00
video_id = self._download_json(
'https://www.telegraaf.nl/graphql', article_id, query={
'query': '''{
article(uid: %s) {
videos {
videoId
}
}
}''' % article_id,
})['data']['article']['videos'][0]['videoId']
2019-11-03 22:04:03 +01:00
item = self._download_json(
'https://content.tmgvideo.nl/playlist/item=%s/playlist.json' % video_id,
video_id)['items'][0]
title = item['title']
formats = []
2019-11-03 22:04:03 +01:00
locations = item.get('locations') or {}
for location in locations.get('adaptive', []):
2019-11-03 22:04:03 +01:00
manifest_url = location.get('src')
if not manifest_url:
continue
ext = determine_ext(manifest_url)
if ext == 'm3u8':
formats.extend(self._extract_m3u8_formats(
2016-07-24 05:29:26 +02:00
manifest_url, video_id, ext='mp4', m3u8_id='hls', fatal=False))
elif ext == 'mpd':
2016-07-24 05:29:09 +02:00
formats.extend(self._extract_mpd_formats(
manifest_url, video_id, mpd_id='dash', fatal=False))
else:
self.report_warning('Unknown adaptive format %s' % ext)
for location in locations.get('progressive', []):
2019-11-03 22:04:03 +01:00
src = try_get(location, lambda x: x['sources'][0]['src'])
if not src:
continue
label = location.get('label')
formats.append({
2019-11-03 22:04:03 +01:00
'url': src,
'width': int_or_none(location.get('width')),
'height': int_or_none(location.get('height')),
'format_id': 'http' + ('-%s' % label if label else ''),
})
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
2019-11-03 22:04:03 +01:00
'description': item.get('description'),
'formats': formats,
2019-11-03 22:04:03 +01:00
'duration': int_or_none(item.get('duration')),
'thumbnail': item.get('poster'),
'timestamp': parse_iso8601(item.get('datecreated'), ' '),
}